One-tenth of a second. That is how long it takes for your brain to decide whether the stranger in front of you is trustworthy, competent, likeable, aggressive, or attractive. Willis and Todorov measured it in 2006: flash an unfamiliar face for 100 milliseconds, and the verdict a viewer renders correlates highly with the verdict formed by people given no time limit at all.
Here is the crueler part. Give someone more time to study your face, and the judgment does not change. It only hardens.
When the researchers stretched exposure from 100 to 500 milliseconds, three things reportedly happened: response times dropped, confidence climbed, and the impressions turned more negative. Push to a full second, and the trait ratings barely moved—but confidence kept rising for some judgments.
Read that sequence carefully, because it is the whole story in miniature. The verdict arrives first. The confidence to defend that verdict arrives second. The extra looking time does not buy accuracy; it buys certainty about a conclusion already reached.
The verdict arrives first. The confidence to defend that verdict arrives second. The extra looking time does not buy accuracy; it buys certainty.
This is not what most people assume happens. We tell ourselves that a longer look means a fairer look—that if we just pay closer attention, we will correct our first read. The data says the opposite. The additional time is arguably spent building a case for the snap judgment, not auditing it.
To understand why correcting a first impression is so hard, you have to see it as a race between two systems moving at wildly different speeds.
The perceptual system is fast. Todorov, reportedly drawing on functional imaging, suspected that judgments of trust involve the amygdala—the brain's ancient threat detector, in place for millions of years before the prefrontal cortex where deliberate reasoning lives.
His phrasing is the load-bearing insight: trust "might be a case of a high-level judgment being made by a low-level brain structure. Perhaps the signal bypasses the cortex altogether."
That is the mechanism. A verdict that feels like a considered assessment of character is, in fact, a low-level structure firing before the reasoning cortex has a chance to weigh in.
Face perception itself moves on the same clock. EEG work reportedly places early face structural encoding (the N170) at 150–190 ms and viewpoint-invariant identity representations (the N250r) at 220–270 ms, though that work concerns the learning of face identity rather than trait judgments. The read is complete in a few hundred milliseconds. Deliberate memory and fact-checking take far longer.
So when you meet someone, your perceptual system has already announced its verdict—danger or safety, competent or not—before the part of you that knows things has finished loading. Correction is not impossible. It is just late.
And there is, reportedly, a directional bias baked into the timing. Recall that as exposure stretched from 100 to 500 milliseconds, impressions grew more negative. As ambiguity resolves, the read arguably tilts toward caution. A brain built to keep you alive would, on this interpretation, rather flag a safe stranger as a threat than the reverse.
Nicholas Rule, who studies snap judgments at the University of Toronto, reportedly built the experiment that turns this from an interesting finding into an uncomfortable one.
Rule and colleagues showed 100 participants photos of 20 men, each labeled as gay or straight. Then they drilled the participants until recall was perfect—every viewer knew, as a memorized fact, each man's stated orientation.
Crucially, half the labels were flipped to contradict how the faces "looked," deliberately setting perception against memory. The point was to force a conflict between what people saw and what they knew.
Then the faces came back, and the time allowed to categorize them was varied.
Under time pressure, participants categorized the men by how they looked—not by the fact they had just memorized perfectly. Given more time, they reverted to what they had been told.
Sit with that. These people were not ignorant. They possessed the correct information and could reproduce it on demand. The bias won anyway, whenever the clock was tight.
Rule's illustration is Ricky Martin, who denied being gay for years before coming out. In the 1990s, he says, you might see the face and think one thing, then recognize the man and remember his own words—"so there's a corrective process there."
His conclusion is the sentence to remember: "First impressions continue to assert themselves long after you know relevant information about a person."
This reframes bias-resistance entirely. The problem is not that people are stupid or bigoted or too lazy to correct themselves. The correct information was sitting right there, memorized, and it still lost.
The problem is a speed asymmetry. Perception delivers first; correction arrives second; and whenever the window is short, first delivery wins. "The less time we have to make our judgments," Rule notes, "the more likely we are to go with our gut, even over fact."
That single line rewires what an intervention should look like. If bias were a matter of attitude or ignorance, the fix would be education and exhortation—teach people the facts, tell them to try harder. But the participants already had the facts and were already trying. Exhortation cannot win a race it enters late.
What can win is time. The only condition under which participants judged by fact instead of appearance was the one where they had enough time for the slower system to arrive.
Which brings us to the practical scandal. Look at where our institutions place their most consequential judgments of other people.
The hiring screen, where a resume photo or a first handshake sets the tone. The courtroom, where jurors form an impression of a defendant's face before a word of testimony—and where research has reportedly long shown "mature" faces draw harsher outcomes than baby-faced ones.
These are precisely the moments we have engineered for speed and single glances. We compress them, we run them on first impressions, we treat the gut read as efficient.
We have, in other words, built the exact conditions under which the corrective system loses.
The fix follows directly from the mechanism, and it is procedural, not attitudinal. You do not fix a timing problem by telling people to feel differently. You fix it by engineering time and structure back into the moment.
That means slowing the judgment down so the slow system can arrive before a decision is locked. It means forcing the relevant facts back into view at the moment of decision, not before it, when perception can still be overruled. It means structured formats over single-glance impressions—rubrics, sequential review, mandatory reconsideration.
There is one further wrinkle worth noting, because it cuts against a comforting assumption. Impressions formed passively—through a video or a Facebook photo rather than in person—are reportedly just as accurate but systematically more negative. Our face-reading is, reportedly, calibrated to live, reciprocal contact; strip that away and the negativity bias runs unchecked. The single-glance photo format is not a neutral shortcut. It is a thumb on the scale.
None of this means first impressions are worthless. Rule's own work reportedly finds people read certain traits at roughly 65 percent accuracy—better than chance, real signal mixed with real noise. The verdict blends true cues with bias, and, arguably, from the inside the two feel identical.
That is exactly why the answer cannot be to trust the gut or to simply distrust it. The gut is fast and sometimes right; the correction is slow and sometimes necessary; and you cannot tell them apart in the tenth of a second when the verdict lands.
So stop asking people to be less biased in the moment. The moment is where they lose. Give the slow system its time, put the facts back in front of the eyes, and let the race be fair.